sb0 changed the topic of #m-labs to: ARTIQ, Migen, MiSoC, Mixxeo & other M-Labs projects :: fka #milkymist :: Logs http://irclog.whitequark.org/m-labs
cyrozap has quit [Quit: Client quit]
cyrozap has joined #m-labs
cyrozap has quit [Quit: Client quit]
cyrozap has joined #m-labs
mithro has quit [Ping timeout: 264 seconds]
mithro has joined #m-labs
sb0 has joined #m-labs
ylamarre has joined #m-labs
ylamarre has quit [Client Quit]
sb0 has quit [Read error: Connection reset by peer]
sb0 has joined #m-labs
fengling has joined #m-labs
fengling has quit [Quit: WeeChat 1.2]
aeris has quit [Quit: en a pas]
sb0 has quit [Read error: Connection reset by peer]
rohitksingh has joined #m-labs
sb0 has joined #m-labs
<cr1901_modern> https://www.kickstarter.com/projects/pine64/pine-a64-first-15-64-bit-single-board-super-comput Something about this being sold for $15 really bugs me
<cr1901_modern> If Raspberry Pi's price point is only possible b/c of Broadcom leverage (i.e. a "normal" SBC project wouldn't have the vendor relationships to negotiate pricing), how can a project claiming to be superior to RPi in every way be able to sell an SBC for half the price?
<sb0> i guess a combination of: do the same with another vendor, cost of chips falling, chip being a relatively small fraction of the cost of those devices anyway
<cr1901_modern> Looks like the RAM alone adds $4 to the cost
<cr1901_modern> if the difference between the 512MB and 1G model is any indication
<sb0> in more exciting news, whitequark was looking yesterday at MPW ASIC runs, which have become really cheap recently
<cr1901_modern> Define "really cheap".
<sb0> raspberry pi et. al. will nicely steal the oxygen of any open source alternative, but for other applications/markets, that can be interesting
<sb0> iirc ~2k USD for a few mm2 of 350nm
<cr1901_modern> Well, the price is definitely going down indeed. Still out of *my* range, but that'll probably be fixed soon enough
<cr1901_modern> I would love to get my own ASICs made, but I don't think I could do it right. I can't even identify transistors when given a picture of a die.
sb0 has quit [Read error: Connection reset by peer]
sb0 has joined #m-labs
<cr1901_modern> John McMaster's site can help me fix that problem I suppose.
<larsc> well you don't have to etch them by hand
<cr1901_modern> larsc: I've been told it's "not as simple as" just placing standard cell library entries onto a die, and that fun analog effects come into play.
<larsc> you probably still have to know what you are doing
<larsc> developing an application is also more than just calling into standard library functions
<larsc> but as things progress custom ASICs will eat into a market section where currently only FPGAs are viable
<cr1901_modern> I've always found studying transistors and how electrons flow through them quite a bit harder than learning how a language maps to assembly instructions XD
<cr1901_modern> I'm just not good w/ materials :(.
<whitequark> fun fact: most engineers don't understand how transistors work
<whitequark> horowitz & hill is quite helpful
<larsc> you don't have to understand how they work, only how they behave
<sb0> oh, that reminds me of a discussion about electric charge and capacitors in #qi-hardware years ago...
<whitequark> larsc: not understanding how they work impedes creating a proper mental model and thus understanding how they behave
<whitequark> sb0: which one?
<sb0> DocScrutinizer preaching that EEs should not know anything about charge
<larsc> whitequark: depends, understanding how it works will allow you to make certain optimizations, but not understanding won't prevent you from using them
<whitequark> well, you don't have to understand how they behave to use them. just copy-paste well-known usage patterns
<whitequark> this happens all too often. i was certainly guilty for many years
<larsc> but then you don't use the element itself, but the pattern
<larsc> but anyway
<larsc> lunch break is over
<cr1901_modern> sb0: Hrm. For circuit models, which are current-based, I'd agree. Definitely not for fabrication tho XD
<sb0> god do i hate airline yield management and those useless RTW tickets
<cr1901_modern> I have two electronics books. Horowitz and Hill for general purpose reference. And Sedra and Smith which is a more detailed reference.
<sb0> useless as in: as expensive as throwaway+hidden city ticketing, except it locks you to one airline alliance plus their booking system proposes you lousy connections so as to reduce the load on normally-booked popular routes
<sb0> those last two points result in ridiculous layovers and unnecessary flights
<sb0> i wish i could just call them and tell them to stop playing their games, they're wasting my time and i'm wasting their seats, everyone loses
<whitequark> i wonder if some kind of travel agent could do exactly that
<whitequark> of course, most of them are just a human interface to hipmunk.com...
<sb0> they won't, because airlines will expel them if they do
<sb0> they also go after passengers sometimes, but that's rare
<cr1901_modern> ^and you know this because... :P?
<sb0> google
<cr1901_modern> Fair enough lol
<sb0> "sort by agony" on hipmunk.com :)
<sb0> hmm, it doesn't show delta, which has the second-fastest HKG->DEN connection, and unlike the fastest (united) has wifi and power on the long haul...
<sb0> (and the duration difference is a dozen minutes)
<sb0> air france are the worst i've seen with that shit, one-ways are more than 3x the price of round trips, they also track your past searches to give you the impression that the prices are continuously rising
<sb0> plus their subsidiary air france consulting gives seminars about this sort of crap to other companies
<cr1901_modern> sb0: Qt 4 compat is required, correct? Is that one reason you always import from quamash?
<cr1901_modern> (quamash seems to be able to in some, but not all, cases, handle incompatibilities between PyQt4 and 5)
<sb0> cr1901_modern, qt4 compat is nice but not an absolute requirement. though as far as i can tell, it is also easy.
<cr1901_modern> I'm not sure why, but in artiq/frontend/artiq_gui.py, you subclass QMainWindow using the QtGui module imported from quamash
<cr1901_modern> When I do the same in my "main" module, I get an AttributeError: PyQt5.QtGui has no attribute QMainWindow
<whitequark> it's in QtWidgets I think
<whitequark> oh right, quamas
<cr1901_modern> It is indeed, in PyQt5
<cr1901_modern> but in quamash, it does some import magic to make it available in QtGui
<cr1901_modern> that is somehow failing in my dummy app...
<whitequark> tbh we'd be better off if we get rid of quamash
<cr1901_modern> It's not required for PyQt to work?
<sb0> cr1901_modern, is it also available in QtWidget from Quamash?
<whitequark> oh, ugh, I misunderstood what quamash did
<whitequark> why the fuck does it poke around in the global Qt namespace?!
<cr1901_modern> sb0: QtWidget works fine
<whitequark> python ecosystem is shit, issue #147
<sb0> cr1901_modern, we should stick to the original Qt namespace in ARTIQ as well...
<cr1901_modern> so why does QtGui work fine in artiq_gui? Hrm...
<cr1901_modern> Alright, that was my follow up q
<cr1901_modern> whitequark: My guess is b/c quamash supports both PyQt 4 and 5, and wants to provide a unified interface
<cr1901_modern> that apparently doesn't work properly
<whitequark> yeah, it should've been split into quamash4 and quamash5
<whitequark> it's not even large
<sb0> cr1901_modern, it definitely doesn't, I have places in artiq where I need to try both imports to support both qt versions
<cr1901_modern> sb0: Okay, so get this bullshit; just for kicks, comment out the pyqtgraph import from artiq_gui.py
<cr1901_modern> Your QtGui imports will fail
<cr1901_modern> (Assuming that Python doesn't choke on the lack of dockarea first)
<cr1901_modern> But adding "from pyqtgraph import dockarea" to my dummy application makes it succeed
<cr1901_modern> So I'm not sure what to do to make it run on both PyQt4 and 5 if the idea is to remove pyqtgraph, or loosen it's influence as a dependency
<cr1901_modern> For convenience, I have uploaded my template main as a gist
<sb0> so importing pyqtgraph meddles with the qt namespaces?
<sb0> ooh is this nasty
<whitequark> ohh, this is what I thought quamash did
<cr1901_modern> whitequark: quamash *does* interfere w/ the Qt namespace: https://github.com/harvimt/quamash/blob/master/quamash/__init__.py#L30-L52
<whitequark> I think only the last two lines do?
<sb0> only l52 does afaict as well
<cr1901_modern> I'm sorry, you're right. Lines 47/49
<whitequark> no
<whitequark> that's not in Qt namespace
<sb0> i have reported the issue with quamash. pyqtgraph is a worse offender, but it has tons of issues open that no one cares about ...
<cr1901_modern> Ahhh, quamash just provides an alternate way to import QApplication that's PyQt-version agnostic
<cr1901_modern> but it doesn't actually *modify* the Qt namespace
<cr1901_modern> Same as QEventLoop, for that matter
<cr1901_modern> (Apparently, PyQt *does* provide it's own EventLoop)
<GitHub72> [artiq] whitequark pushed 2 new commits to master: http://git.io/vRjdP
<GitHub72> artiq/master 22dab0c whitequark: compiler: clarify confusing comment (NFC).
<GitHub72> artiq/master ccc9930 whitequark: transforms.artiq_ir_generator: treat builtins in 'with' syntactically.
<whitequark> sb0: the watchdog function accepts a value in seconds, right?
<whitequark> there's no watchdog_mu?
<whitequark> (should there be?)
<sb0> whitequark, yes, seconds
<sb0> the runtime function takes milliseconds
<sb0> we don't need high precision there
<whitequark> ok
<GitHub150> [artiq] whitequark pushed 6 new commits to master: http://git.io/v0ePs
<GitHub150> artiq/master 3ec9b67 whitequark: Revert "transforms.artiq_ir_generator: treat builtins in 'with' syntactically."...
<GitHub150> artiq/master c660028 whitequark: transforms.llvm_ir_generator: fix reraise inside EH block.
<GitHub150> artiq/master 143bae4 whitequark: compiler.builtins: {fn→obj}_{serial,parallel}.
<bb-m-labs> build #37 of artiq is complete: Failure [failed lit_test] Build details are at http://m-labs-buildserver.lan/buildbot/builders/artiq/builds/37
<whitequark> grumble
<GitHub165> [artiq] whitequark pushed 1 new commit to master: http://git.io/v0ey6
<GitHub165> artiq/master 25027f6 whitequark: compiler: fix tests.
<bb-m-labs> build #38 of artiq is complete: Success [build successful] Build details are at http://m-labs-buildserver.lan/buildbot/builders/artiq/builds/38
kyak has quit []
ylamarre has joined #m-labs
kyak has joined #m-labs
kyak has joined #m-labs
kyak has quit [Changing host]
<sb0> of course, the pyqtgraph dock size management code is completely dysfunctional, which is why some artiq popup windows have grossly inappropriate sizes right now
<GitHub103> [artiq] sbourdeauducq pushed 2 new commits to master: http://git.io/v0vld
<GitHub103> artiq/master c4b33bd Sebastien Bourdeauducq: examples/speed_benchmark: fix nruns step
<GitHub103> artiq/master 0eb8079 Sebastien Bourdeauducq: gui: fix some dock sizing issues
ylamarre has quit [Quit: ylamarre]
mumptai has joined #m-labs
ylamarre has joined #m-labs
ylamarre has quit [Client Quit]
mumptai has quit [Quit: Verlassend]
rohitksingh has quit [Quit: Leaving.]